【作者】 吴琼；

【导师】 杨以涵； 刘文颖；

【作者基本信息】 华北电力大学（北京） ， 电力系统及其自动化， 2008， 博士

【摘要】 确保电力系统的安全、稳定、经济、优质运行是调度人员的主要职责。电网规模不断扩大和结构日益复杂,使调度人员面临着巨大挑战。然而,作为主要辅助决策工具的能量管理系统(EMS, Energy Management System)并不能完全满足新环境对电网运行监控的要求,其应用范围和决策的智能化水平都有待提高。基于以上背景,本文提出建立功能完善、使用方便、可以在线自主协调运行的“电网智能调度决策支持系统”的设想。电网智能调度决策支持系统将更多的智能技术应用于电力系统分析,为调度分析与操作提供智能化的决策支持。本文通过对电力系统暂态稳定预测和电网操作指导的研究,进一步完善了智能决策支持系统的分析功能,并采用多智能体技术使系统具有在线自主协调运行能力,从而可以为调度人员提供更为全面的决策支持。研究内容分为两个层次:一是决策分析工具的扩充,二是系统自主运行和协调决策能力的实现。考虑到电力系统安全稳定运行的重要性,在充分利用广域测量数据的基础上,提出了一种基于在线式学习算法的暂态稳定预测方法。该方法根据分块矩阵求逆定理对标准学习算法进行改进,以提高计算速度。为满足实际多机系统稳定预测的要求,引入轨迹聚合技术对多机轨迹进行聚合,进一步减少了计算量。在轨迹降阶的基础上,根据扩展等面积法则(EEAC, Extended Equal Area Criteria),通过识别聚合轨迹的动态鞍点来判断轨迹的稳定性。仿真试验结果从预测精度和计算时间两方面验证了方法的有效性。在电网操作指导方面,深入分析了操作票专家系统中广泛采用的产生式规则表示方法的局限性,指出推理程序对规则表示的依赖是制约操作票专家系统通用性的根本原因。为解决上述问题,首先研究了产生式规则的Petri网表示方法,提高规则表示的灵活性;其次,研究了基于Petri网模型的推理方法,使推理程序与规则表示完全分离,从而提高了操作票系统的适应能力。为了实现软件系统的自主运行,将多智能体技术引入调度决策支持系统的研究中。通过划分电力系统运行状态和定义典型事件,建立了电力系统领域环境模型,并以此作为软件智能体的生存环境,在识别电力系统运行状态和事件的基础上赋予智能体环境感知能力,从而解决了软件自主运行问题。通过研究基于环境交互的多智能体协作方法并将其用于调度决策方案的在线自动生成,使多个软件智能体可以协调运行,主动为调度人员提供针对电网当前运行状态的调度决策方案。更多还原

【Abstract】 Because of the expending scale and complexity of power grid, the power system operator is facing with much greater challenge the duty of ensuring the system’s security, stability, economy and operating quality. However, the energy management system (EMS) which serves as a major tool to support decision making can’t fully meet the requirement of power system operation monitoring, its function and intelligent level need to be improved. Under this background, this paper presents a notion of constructing an intelligent decision support system for perfect function, convenient using and online self-coordinated operation. More AI techniques are used in the intelligent decision support system to provide better decision results to the operators. Based on the research of power system transient stability prediction and switching sequence generation of power grid, this paper improves the analysis function of the intelligent decision support system, realizes online coordinated operation capacity by adopting multi-agent technology. By such means the system can provide integrated decision support for the operator. This paper includes two hierarchies: the expansion of the decision-making analysis tools and the realization of system autonomous running and coordinated decision-making capacity.Considering security and stability of power grid as the primary goal, this paper presents a transient stability online prediction method based on machine learning algorithm. According to the theorem of inverting block matrix, the standard machine learning algorithm is improved, and the learning efficiency is enhanced. In order to satisfy the online transient stability prediction of multi-machine system, polymerization technology of trajectories is used to reduce the computing burden. Based on the theory of extended equal area criteria (EEAC), transient stability of power system is estimated by identifying the dynamic saddle point (DSP). According to the simulation results, the validity of proposed method is proved in respects of prediction precision and computing time.As to switching sequence generation of power grid, this paper analyses the disadvantage of production knowledge representation which has been widely used in operating command expert system, and indicates that dependent relation between inference program and rules is the main reason that restricts the adaptability of operating command expert system. To solve the above problem, firstly, this paper establishes a more flexible model of production knowledge based on Petri net; secondly, a new inference engine is brought forward, which can separate inference program from knowledge representation and improve adaptability of operating command expert system.In order to realize system autonomous running, this paper introduces multi-agent technology into the research of the intelligent decision support system. By defining power system operation states and typical events, the field model that agents live in is constructed. Based on the recognition of operation states and events, this paper granted sensation capacity to intelligent agents. By the research of multi-agent collaboration based on interactive theory and applying to the dispatching schemes auto-generation, the system can actively provide decision results for power grid operator according to current power grid operation state.更多还原